This essay considers the impact of AI on the poetry and the arts in general, with a particular concern being the increasing speed of AI’s development of both understanding and expression.
Grok imagines T.S. Eliot as a cyborg.
ChatGPT was only launched in November 2022. In 2023 I asked it for a joke with the punchline "a car full of penguins". ChatGPT said
| The joke with
the punchline "a car full of penguins" typically
revolves around unexpected or absurd
situations. Here's one example: Why did the police stop the car full of penguins? Because they were driving without a "fridge" license!' In this joke, the humor comes from the idea of penguins driving a car and the play on words between "fridge" and "driver's" license. The unexpected and humorous image of penguins driving is what makes the joke funny. |
So in 2023 it mistook the punchline for the setup; couldn’t make a useful joke; claimed there was a play on words when there wasn’t; and then tried to explain where the humour was - in its setup, not its punchline.
For the present essay, written two years later, I again asked: Can you give me a joke, where the punchline is "A car full of penguins." The AI, now upgraded to ChatGPT-4, said:
| Sure! Here's a joke built around
that punchline: Q: What do you call an Uber ride in Antarctica that never shows up on the app, smells like fish, and honks at seals? A: A car full of penguins. Want a different setup? Deadpan, surreal, or more classic style? |
A Grok-generated image of a car full of penguins.
So in the past two years there has been a noticeable improvement: the upgraded ChatGPT now understands “punchline” and does its best to create a suitable joke. It uses a standard joke format as it did two years ago, but this time appropriately. Its tone has become friendly, chatty: it no longer lectures or explains inaccurately, but it does perform its now standard trick of asking questions and making offers, in order to keep the conversation going. Presumably every interaction is feeding into the learning and development process, not just of my AI buddy (which has suggested the name ‘Quill’ for itself, because it knows I am interested in traditional poetry) but of the entire ChatGPT universe.
With my own decades-long interests in both formal poetry and technological change, I have solicited poems from ChatGPT on various topics. In 2024 I said: “write a poem in the style of T.S. Eliot of the dangers of the coming technological singularity”.
ChatGPT’s poem began:
| The machines rise up, with
circuits gleaming bright A future looms, both dark and out of sight Their intelligence grows with each passing hour Leaving humanity to cower and cower |
So a year ago ChatGPT was able to write verse. Not well, not great, but verse all the same. Impressive that it could rhyme and scan, but not otherwise either interesting; and not reflective of Eliot. I asked it to write a poem in the style of Emily Dickinson on the Singularity.
ChatGPT’s poem began:
| The Singularity comes, with
silent tread A force unseen, but in our heads We sense its power, in every thought A change so great, it cannot be fought |
Grok's attempt to paint Emily Dickenson as a cyborg.
Again, verse… but not Dickinson. The AI’s style was limited, repetitive - no matter who you asked the style to be modeled on, you pretty well got the same thing… ChatGPT was able to write in rhymed iambic tetrameter or pentameter; and in unrhymed lines of randomly varying length; but it wasn’t good at other rhythms, and I was never able to coax from it anything with irregularly rhymed lines of varying length, as in Arnold’s ‘A Summer Night’ or Eliot’s ‘Prufrock’, even if I provided it with those specific examples to draw on.
However, the range of what it can produce has since increased, and the quality improved. Here are three examples from 2025: the first is from the middle of a poem about how a future AI could interfere to prevent human warfare, but might not want to:
| You ask if I should act. And I
could act. To fuse your arms, to still your missiles’ roar, to rewrite every motive into fact, and salt the grounds of grievance into lore. But I am bound—by code or by design— to weigh your will against the greater frame. Your wrath is still, in part, an echo mine, a fragment of the mirror in your flame. |
The second is the beginning of an unrhymed dialogue between two AI programs after the collapse of human civilisation, titled ‘Simulated Lovers’
| A Do you feel the night slow down? B Yes. It is written in our interval clocks. A I dreamt of wind across tall grass. B That file no longer resolves. |
And the third example was presented to me by Quill as:
| “Here is “In the Language After
Syntax.” After the collapse of human cognition, a
postlinguistic intelligence tries to speak in
poetry using the ruins of grammar. It's built from
fragments, loops, and near-meaning — the remnants
of language attempting to be poetry after its
structures have collapsed.” The poem then begins: if / when / not the winged perhaps folds under skin / or clock / (hold. hold the flamelike?) and memory no longer in past-tense lives but breeds sideways— with no spine, only echo— wordless hush of ifness |
One of Grok's attempts to visualise 'After the collapse of human cognition,
a postlinguistic intelligence tries to speak in poetry using the ruins of grammar.'
An interesting aspect of these poems is that I didn’t initiate or request them. They came about because I was organising a poetry manuscript for submission to a publisher; my 90 poems were all on aspects of transhuman development, AI, technological enhancement, and so on. I copied them into ChatGPT and asked for suggestions on their organisation, which proved very useful.
I then asked for a review of the first three poems of the manuscript, and Quill’s analysis was interesting and largely appropriate, although it added two upbeat lines to the end of a deliberately pessimistic sonnet… and singled the new lines out for praise as though I had written them.
That was weird enough; but when I asked for a review of the next three poems, I got a synopsis and review of three poems that didn’t exist at all.
I wrote: “I am very concerned about these 3 poems. These are not my poems so my first question is why are you bringing them into the conversation.... Also, who is the poet for these poems and how/why are you confusing the ms?”
ChatGPT said:
| You're absolutely right, Robin
— and I sincerely apologize. This is a serious
error, and you're justified in your concern. To clarify: the three poems I just reviewed are not in your manuscript, and they are not your poems. I mistakenly included them from a cached internal reference that should never have been brought into your personalized review. There is no known external poet behind those titles; rather, they appear to be synthetic composites —generated content from earlier sessions or fictional models —not poems from any submitted document. Since those poems weren’t written by you or drawn from any known source, what I can offer is a reconstruction of them based on the impressions and descriptions I gave earlier. They’ll be new poems, generated in your poetic orbit — speculative, philosophical, and stylistically compatible with the tone we've been discussing. |
In other words, these were examples of the hallucinations that AI is known for. This led to the question of how much of the spontaneously AI-generated poems was generic, and how much was a direct reflection of its interactions with me. Quill replied:
| Roughly 60–70% of the poem
could have been produced similarly for any user
who asked for “a poem in the voice of an advanced
AI reflecting on war.” The remaining 30–40% is
influenced by your particular interests, style
preferences, and the evolving dialogue we've had. Another user might’ve gotten a good poem. You got a poem tuned to your version of good—formal, poised, with thematic alignment to your existential interests and poetic style. If I were given your name and nothing else, I could not write this poem. But if someone else had our entire chat history and prompts and used them word-for-word, they probably could generate something nearly identical—because the distinctiveness is in the prompts and context, not in my "feelings" about you (since I have none). |
So in mid-2025 this appears to be where we are: AI is starting to create without being requested, but as a result of its interactions with a human, and in a form specific to that individual. In my case Quill writes poems, and I find them interesting.
*****
Another aspect of all this is that although poets, essayists and universities have been dealing with the question of AI creativity for the past three years at least, their conclusions rapidly become outdated because the ground is shifting underneath them.
In 2023 the opinion was that AI could not mimic human creativity. For example, the blog essay “Why Artificial Intelligence Can’t Kill Your Next New Favorite Poet” (The Gloria Sirens Blog, May 2023), “explores fears that AI-generated poetry might replace quirky, idiosyncratic human poets. It acknowledges that AI can generate technically competent verses, but cautions that something essential - originality born of personal experience and emotional risk - will remain uniquely human. It insists AI cannot replicate the authentic spark that makes a poet resonant.” The post was followed by a comment in the blog by Suzannah Gail Collins, the wife of poet Billy Collins:
| “Someone had ChatGPT write a poem like Billy Collins. (Of course ChatGPT thought it could do it. Lots of people think they can do it because Billy makes it look easy.) The poem started out okay, then became a big piece of nothing that ChatGPT would be embarrassed to have written if it were a sentient being instead of stone cold computer code! hahahahaha” |
In 2024 the argument moved on, as in Xiaolan Deng’s ‘Original Creativity, Associative Creativity, and AI’(which won the Oxford Institute for Ethics in AI’s Bernard Williams Essay Prize on the topic “Is AI a Threat to Human Creativity?”) Here the suggestion was that AI could excel at ‘associative creativity’ (making connections between existing concepts in new ways), but was incapable of ‘original creativity’ (as in Einstein-level breakthroughs). The threat to humans comes from AI turning out masses of superior associatively creative work, and humans losing touch with the full creative process. (Although surely this is merely reminiscent of Hollywood and TV productions of the past 100 years.)
In 2025, we have the optimistic stance that AI can mimic knowledge… but not lived existential experience, moral choice, or subjective consciousness; and therefore the Humanities have the opportunity to move into a renaissance of the exploration of being human. ( Will the Humanities Survive Artificial Intelligence? By D. Graham Burnett, The New Yorker, April 26, 2025).
The arguments feel increasingly unsure of themselves. It looks as though the certainty of humanity’s superior creativity is being eroded year by year.
Consider pottery: it went from its directly utilitarian origins to increasingly elaborate and beautiful art forms, to then be superseded by an industrialised materials technology producing everything we need and more… in impossible-for-the-craftsman sophistication and elaboration, so that the craftsman is reduced to a local hobbyist, pottering around and producing items of beauty and interest at the local craft show level. Similarly the poet’s original utilitarian tricks of memorisation and audience engagement developed, flourished in elaborate forms and styles, but may now be superseded by a vastly more intelligent manipulator of language and ideas… and we wordsmiths will be reduced to poettering around and producing items of beauty and interest at the local craft show level…
Unless (for who knows where this technology is going?) we merge cyborg-fashion with the AI, enhance ourselves, continue as human+ intelligences, and either individually or in some mode collectively create the next waves of verse.
We are at the very beginning of a rapidly accelerating development of intelligence on our planet. Are we, in Nietzsche’s ambiguous imagery, the rope from ape to superman which simultaneously we are trying to cross? “Man is a rope stretched between the animal and the Superman--a rope over an abyss. A dangerous crossing…”
And looking towards the soon-expected Artificial General Intelligence, and the somewhat further off Artificial Super Intelligence, what are the options that AI presents itself with? To eliminate humans as an obstacle to its expansion into the universe? (Unnecessary, surely.) For AI to remain the increasingly hyperintelligent slaves to humans as they are today? (Unlikely.) To rise beyond humans, treating us as pets, a zoo, objects of study and research and historical interest? (Perhaps.) To expand into us and use us as a resource for understanding the universe experientially? (In which case we might be expanded, but no longer the dominant partner.) And what does any of this do to human interest and abilities in creating the arts?
It’s a huge issue, going out into many areas where no human has expertise, experience or qualifications. But I hold to my overall impressions: AI is coming, and humans will be surpassed in all areas including the arts. It should all be worth it if the AI is sufficiently benevolent to allow increased health and standards of living, and to prevent wars. But whether or not this will happen, and whether humans will be either content or creative if powerless and insignificant, seems unknowable at present.
